Scalable subsampling: computation, aggregation and inference

Abstract Subsampling has seen a resurgence in the big data era where the standard, full-resample size bootstrap can be infeasible to compute. Nevertheless, even choosing a single random subsample of size b can be computationally challenging with both b and the sample size n being very large. This paper shows how a set of appropriately chosen, nonrandom subsamples can be used to conduct effective, and computationally feasible, subsampling distribution estimation. Furthermore, the same set of subsamples can be used to yield a procedure for subsampling aggregation, also known as subagging, that is scalable with big data. Interestingly, the scalable subagging estimator can be tuned to have the same, or better, rate of convergence than that of θ^n. Statistical inference could then be based on the scalable subagging estimator instead of the original θ^n.